The paper has two main goals. Firstly, it is aimed at proving that there is a single principle governing the choice of any portion of the original – from morpheme to the entire text - for the role of the unit of translation (UT). It is based on the role a linguistic unit plays in the bigger form of which it is an integral part. If it makes its individual input into the meaning of the whole then it should be given special attention in translation, i.e. at some stage of re-coding made a UT. But if the meaning of the whole larger construction is such that it is not made up by putting together the meanings of its composite pars – a situation termed idiomatic – then (and only then) the entire whole is taken as a unit of translation. The paper also shows that when theorists declare that there is only one linguistic entity which can be qualified as a UT – in some works this is the sentence, in others, the entire text – they are using the term in their own interpretation and not in the meaning that was give the term by its authors. But it is essential for any theory that its terms, in the case discussed – the term ‘unit of translation’ - be applied by all in one and the same meaning, and exactly in the meaning that was given it at inception. Because the UT was defined as a portion of the original text, it would seem that the text as a whole cannot serve as a UT. The second aim of the paper, however, is to show that there are certain types of texts that answer the same requirements for serving as a UT that are valid for all other linguistic units. These are poetical texts of such fineness that places them onto their own highest level of linguistic structures. It is these texts that for purposes of re-coding demand being taken as a whole, i.e. as an undivided unit of translation.